Children’s Eye Movements during Reading (стр. 2 )

Third, to reiterate, there are several well-developed computational models of eye movements during reading but they are based solely on data from skilled adult readers, (Engbert et al., 2002; Engbert et al., 2005; Pollatsek et al., 2006; Reichle et al., 1998; Reichle et al., 1999, 2003; Reilly & Radach, 2006). Recent work using artificial learning agents has shown how eye movements resembling those of a skilled adult reader can emerge as a consequence of reinforcement learning (Liu & Reichle, 2010; Reichle & Laurent, 2006). No a priori assumptions were made about the mechanisms underlying eye movement control; the agents were simply limited by constraints relating to factors known to impact on eye movements during reading (time taken to program saccades, word length as an index of processing difficulty associated with lexical identification, visual acuity limitations on lexical identification, and distribution of attention across words in a sentence). The agent was rewarded for each word that was identified, and was punished for time spent on word identification – the goal was that all the words in the sentence should be identified as quickly as possible. The behaviours that emerged were similar to those observed in skilled adult reading. The agent learned to target saccades toward the centres of words, and decisions about when (reading times) and where (refixation and word skipping probabilities) to move the eyes were affected by the processing difficulty associated with a word. In the future, modelling eye movement data collected from children as they read will provide the opportunity to understand the mechanisms underlying developmental changes in eye movement behaviour during reading, such as vision, language, attention, and basic oculomotor control, and interactions between such mechanisms during the progression from beginning to skilled reader.

2.2. Methodological issues.

There are several methodological issues that should be considered in relation to developmental studies of eye movements during reading. The first issue is that of control groups, and this relates to the theoretical issue outlined above of how both chronological age and reading skill might contribute to the development of eye movement behaviour during reading. In the small number of studies which have used different groups of children in order to examine age-related changes in reading performance, groups have generally been split by chronological age. While this has certainly proven fruitful, it is worth noting that in other areas of developmental research (such as investigation of the language and literacy difficulties associated with developmental disorders), comparison groups are often controlled both for chronological age and for cognitive ability or mental age (e. g., Bishop & Snowling, 2004; Laing, Hulme, Grant, & Karmiloff-Smith, 2001; Leonard, 1998; Snowling, 2000). It is important that research is undertaken to directly address the relative contributions of age and reading skill to the development of eye movement behaviour during reading. Until this is better understood, the inclusion of control groups for both chronological age and reading age would be helpful to avoid inadvertent confounding of one with the other.

A second methodological issue worth noting is the importance of the use of age-appropriate linguistic stimuli. One factor distinguishing between the studies reviewed in this chapter is their selection of stimuli for readers of different ages. Some studies have presented each group with age-appropriate stimuli. In this case, processing difficulty associated with reading can be equated across the groups (assuming careful stimulus selection), but the analysis of age group effects will necessarily be between-items; differences between groups may be attributable to their age, or may be attributable to the particular set of sentences that was read by each group. In other studies, participants of all ages read the same stimuli. In this latter case, analyses of age group effects are within-items, thus eliminating the concern that any effects observed are related to differences in materials rather than reflecting genuine effects of age group. There is, however, the concern that any set of reading materials will be inherently more difficult for younger readers compared to older readers. It is well-documented that for skilled adult readers, increased processing difficulty impacts on eye movement behaviour. For example, word reading times are longer on low frequency words than high frequency words (e. g., Henderson & Ferreira, 1990), and reading times are longer and more regressions are made for sentences that are syntactically complex compared to those that are simple (Warren, White, & Reichle, 2009); for a summary see Rayner (1998, pp. 389-392). Clearly, having participants of different ages read the same materials will lead to differences in eye movement behaviour that are, at least partially, attributable to the processing difficulty associated with the text for readers of different ages.

A related point is that manipulations of text difficulty may affect readers differentially depending upon their reading skill and experience of processing printed text. Factors known to have a robust effect on adults’ eye movements during reading, such as the length or frequency of a word, need to be carefully considered before being incorporated into experimental designs with children. For example, there are many words which are low frequency for adult readers but high frequency for children (e. g., dragon) and frequency of encounter, together with cumulative frequency (the total number of times a word is encountered over time) and Age-of-Acquisition (the age at which a word was acquired), may impact differently on lexical retrieval processes in adults and children. Ideally, future research would control for age-appropriateness by constructing multiple sets of experimental sentences, each designed for a different age group (e. g., Häikiö et al., 2009) and, if possible, by using extensive pre-screening in order to compose sentences that are roughly equivalent in processing difficulty for all participant groups.

While creating stimuli that include linguistic processing manipulations but are also comparable in terms of processing difficulty across different age groups is extremely challenging, in certain circumstances it may be possible. For example, it is possible to set up a frequency manipulation in which items are high frequency for both adults and children of different ages. In addition, context can be made particularly relevant to children (e. g., sentences refer to well-known children’s characters, school etc.), thereby reducing (although not eliminating) the gap between how easy adults and children will find sentences to process. In addition, stimuli can pre-screened for variables such as comprehension, plausibility, and predictability. Experimenters can attempt to equate these across age groups by pre-screening more items than are needed and selecting a subset for the main experiment that are matched on these variables between adults and children.

Third, there is a methodological issue which inevitably arises when conducting research with children – increased variability in data. This usually exists not only in eye movement measures, but also in related measures such as reading ability, and can lead to problems with statistical analyses (e. g., violating the assumption of equal variance across groups for ANOVAs) as well as making it difficult to obtain reliable effects. One possible solution is to use large sample sizes. Alternatively, child participants may be grouped more closely with respect to either chronological age or reading skill. Both solutions are increasingly possible as new eye tracking devices make it far easier to test large numbers of children. Finally the increasing use of Linear Mixed Effects (LME) analyses in eye movement research, in which both fixed and random effects are accounted for, may reduce the problems caused by this increased variance in child data. It may, however, also be the case that it is not realistic to expect reliable differences when experimental manipulations which generate only small effects are employed. Certainly, eye movement research into children’s reading is at an early stage compared to the advanced literature on skilled adult readers where increasingly complex manipulations are employed to investigate relatively small effects. Currently, the basic questions that need addressing with respect to children’s eye movements during reading allow studies to employ strong manipulations that generate large differences between groups of children. In the future, however, the issue of increased variability in children’s data may be addressed by more sensitive approaches to the selective recruitment and screening of child participants, together with careful design and screening of experimental stimuli.

3. Basic characteristics of children’s eye movements during reading

For comparative purposes, means for different eye movement measures at different ages from all of the published studies to date of typically developing children’s eye movements during reading are shown in Table 1. The age-related changes are quite clear; as chronological age increases, sentence reading times and fixation durations decrease, saccade amplitudes increase, fewer fixations and regressions are made, refixation probability decreases and word skipping probability increases (Blythe, Häikiö, Bertam, Liversedge, & Hyönä, 2010; Blythe et al., 2006; Blythe, Liversedge, Joseph, White, & Rayner, 2009; Buswell, 1922; Häikiö et al., 2009; Huestegge et al., 2009; Joseph, Liversedge, Blythe, White, & Rayner, 2009; McConkie et al., 1991; Rayner, 1986; Taylor, 1965; Taylor, Frackenpohl, & Pettee, 1960). While all studies show these same developmental changes in eye movement behaviour, they vary in their selection of stimuli for readers of different ages. Some studies have used different materials for readers of different ages in an effort to match reading difficulty across groups (Blythe et al., 2006; McConkie et al., 1991; Taylor, 1965). In contrast, other studies have presented readers of all ages with the same sentences in order to avoid differences in material as a confounding variable between groups (Blythe et al., 2010; Blythe et al., 2009; Buswell, 1922; Häikiö et al., 2009; Huestegge et al., 2009; Joseph et al., 2008; Joseph et al., 2009; Rayner, 1986). In this second group of studies, the stimuli were typically aimed at the level of the youngest readers tested and, as a consequence, were very easy to read for the adult participants.